Author Affiliation

Department of Computational Modeling and Simulation Engineering, Old Dominion University

Publication Date

2022

Conference Title

Modeling, Simulation and Visualization Student Capstone Conference 2022

Conference Track

Medical Simulation

Document Type

Paper

Abstract

There are many issues regarding the technology used for fall detection gait analysis in the geriatric center of senior patients. The fall detection system used should examine the privacy of the patients and the flexibility of the system. Several researchers have developed fall detection using wearable sensors due to their flexibility and nature of privacy. Most of those developed methods are supervised deep learning methods. However, data annotation is expensive because we use camera video recording and playback of each participant’s recorded video to label the data. Moreover, labelling using a camera recording limits the flexible and private nature of wearable sensor-based fall detection. This paper presents how to use unlabelled data to pre-train our models and use labelled data to fine-tune those pre-trained weights. We collected unlabelled and labelled data and applied self-supervised learning to detect falls. First, we performed pre-training on the unlabeled data using ResNet model. After that, fine-tune and train ResNet using the labelled dataset. The experiment in this study suggested that the best performance can be achieved by using pre-trained weights of unlabelled data from the accelerometer and gyroscope sensors. Furthermore, oversampling and modified loss functions are used to handle the dataset’s imbalance classes. With the ResNet pre-trained weights and re-training using the labelled data, the experiments achieved an F1-Score of 0.98.

Keywords:

Gait analysis, Fall detection, Senior subjects, Self-supervised learning

Start Date

4-14-2022

End Date

4-14-2022

DOI

10.25776/p8n9-q949

Share

COinS
 
Apr 14th, 12:00 AM Apr 14th, 12:00 AM

Senior Subject Gait Analysis Using Self-supervised Method

There are many issues regarding the technology used for fall detection gait analysis in the geriatric center of senior patients. The fall detection system used should examine the privacy of the patients and the flexibility of the system. Several researchers have developed fall detection using wearable sensors due to their flexibility and nature of privacy. Most of those developed methods are supervised deep learning methods. However, data annotation is expensive because we use camera video recording and playback of each participant’s recorded video to label the data. Moreover, labelling using a camera recording limits the flexible and private nature of wearable sensor-based fall detection. This paper presents how to use unlabelled data to pre-train our models and use labelled data to fine-tune those pre-trained weights. We collected unlabelled and labelled data and applied self-supervised learning to detect falls. First, we performed pre-training on the unlabeled data using ResNet model. After that, fine-tune and train ResNet using the labelled dataset. The experiment in this study suggested that the best performance can be achieved by using pre-trained weights of unlabelled data from the accelerometer and gyroscope sensors. Furthermore, oversampling and modified loss functions are used to handle the dataset’s imbalance classes. With the ResNet pre-trained weights and re-training using the labelled data, the experiments achieved an F1-Score of 0.98.